YuhaoWong0103 / fsl_ts

few shot learning (MAML) for time series prediction
MIT License
9 stars 4 forks source link

About paper or explanation about this project #1

Open 17011813 opened 2 years ago

17011813 commented 2 years ago

Good Morning!

Can you explain your project in more detail about time-series dataset that you used in this project? I wanna understand your project more specifically :) Is this classification or regression task? Which task is classification or regression?

image

Thank you~!

YuhaoWong0103 commented 2 years ago

Hello there, thanks for asking!

The time-series dataset involved in this project comes from the KPI data collected in the wireless network. It's a regression task on time-series prediciton. :)

17011813 commented 2 years ago

Thanks for answering my questions!!

Where did you get original time-series datasets? And how did you preprocess the dataset? Do you have some code about this?

I want to execute your code with my own time series dataset!

Can you give me some advice about this issue?

Thank you so much :)

YuhaoWong0103 commented 2 years ago

The dataset is provided by the enterprise and is not open source, so I can't share the data preprocessing code with you, unfortunately.

But I can describe the basic shape of the dataset. I hope this helps you with your experiments. I use a sliding window to divide each original time series into several sub-time series, which corresponds to the last dimension of the shape of the dataset (the length of the sub-time series is 12).

The total shape of the dataset is (task batch_size, support_size / query_size, the number of sub-time series, the length of sub-time series)

licj1 commented 2 years ago

Good Morning!

Can you explain your project in more detail about time-series dataset that you used in this project? I wanna understand your project more specifically :) Is this classification or regression task? Which task is classification or regression?

image

Thank you~!

Hello, I'm also reproducing the code recently, is it convenient to discuss it together? My email is: lcjlcj666888@gmail.com